作者：马纯北安 来源：原创 发布日期：08-24
一 | 记者24日从兰州大学获悉，该校物理科学与技术学院田永辉教授团队在光学数字计算领域取得重要进展。该团队研制出一种可重构光学算术逻辑单元芯片，将多种逻辑与算术运算集成于同一硬件平台。相关成果发表于国际权威期刊《自然·通讯》。 (ECNS) -- Chinese humanoid robot maker Unitree Robotics has attracted strong market attention after its IPO subscription results showed the lowest winning rate in the history of Shanghai’s STAR Market. Unitree priced its shares at 150.80 yuan ($22.2) each, requiring successful applicants to pay 75,400 yuan for one lot of 500 shares. The online winning rate was just 0.0181 percent, prompting comparisons to “winning a lottery.” Unitree Robotics issued about 40.45 million shares, accounting for 10 percent of its post-IPO total shares, and is expected to raise about 6.1 billion yuan. Based on the issue price, the company’s market capitalization is estimated at about 61 billion yuan. The IPO attracted a strong lineup of strategic investors, led by the National Council for Social Security Fund. Major technology and industrial investors also participated, including DeepSeek, Kunlun Capital under China National Petroleum Corporation, China Southern Power Grid and China Telecom Capital. DeepSeek received about 933,000 shares worth 141 million yuan, subject to a 36-month lock-up. Meituan-affiliated investment entities are among the company’s largest external shareholders, while Tencent, Alibaba and Ant Group also hold stakes in the company. Leading venture capital firms Sequoia China and Matrix Partners China, along with state-backed investors such as the Beijing Robotics Industry Development Investment Fund and Shenzhen Capital Group, further broadened support. Founder Wang Xingxing is also expected to benefit from the listing. Before the IPO, Wang held about 33.36% of the company’s shares directly and indirectly. Based on the issue price, his stake was valued at around 18.3 billion yuan, though his ownership ratio will be diluted after the listing. The IPO also lifted related robotics stocks on Aug. 11 and 12. (By Helen Mo, intern Liu Shuangjing) 
。 随着人工智能与数据密集型应用对高速、低时延计算需求的激增，传统电子集成电路正面临摩尔定律放缓、互连带宽受限和功耗持续上升等挑战。集成光子技术凭借大带宽、高速并行处理和低传输时延等优势，为突破电子计算瓶颈提供了重要的技术路径。 基于这一背景，研究团队依托成熟的硅光子集成平台，利用微环调制器体积小、调制速度快、功耗低且易于电控重构的特点，研制出可实现多种逻辑与算术功能动态切换的光学数字计算芯片。团队通过级联微环构建基本逻辑单元，实现了NOT、AND、OR、XOR和XNOR等多种逻辑运算；同时引入波长维度并行处理不同二进制位，使同一硬件架构能够动态切换为2比特加法器、减法器和数值比较器，无需重新设计芯片结构。 芯片基于220纳米硅绝缘体平台制备，每个微环集成载流子耗尽型高速电光调制结构和热调谐单元，可分别完成高速数据加载与工作波长校准。实验结果表明，该系统在20Gbit/s速率下实现了基本逻辑运算及算术功能，计算密度达到528Gbit/s2，能效达12.36fJ/bit。 在此基础上，团队完成了高速数据加密与解密、医学图像水印嵌入和图像差异增强等应用验证，展示了该架构从基础逻辑器件向多功能光学数字计算系统扩展的能力。 田永辉表示，该工作为构建高密度、低功耗、可重构的光子算术逻辑处理芯片提供了新的实现路径，也为光学数字计算在高速信息处理和片上智能计算中的应用奠定了器件基础。（记者颉满斌 通讯员余若怡）。
Current article:http://rojn61.naoninanglouzhei.bond/list_lyzh/neok.html
Published on:07:51:29